South Korea is about to run a fascinating experiment in digital democracy. On June 3, the country holds local elections — and for the first time, two laws targeting AI deepfakes in politics will be fully enforced. One bans realistic fake audio or video for 90 days before election day, with prison time and heavy fines for violations. The other requires businesses to clearly label any AI-generated content, including through watermarks. The big question everyone is asking: will it actually work?
Deepfakes have been around for years, but their quality and access have increased alongside the sophistication of generative AI. What once required serious technical skill can now be done in minutes with consumer apps. South Korea has already seen fake videos of candidates go viral, and a fabricated clip falsely showing a presidential frontrunner ending his hunger strike spread widely before anyone could stop it. The speed is the problem. By the time a platform removes a clip, it has already reached the people it was designed to reach.
South Korea does deserve credit for building real legal infrastructure around this issue — something most countries haven't managed. The National Police Agency even developed its own deepfake detection tool. But experts warn that regulation only sets a floor, not a ceiling. Threat actors operating from outside the country face no practical liability, and encrypted messaging apps sit beyond the reach of most regulators.
For comparison, the US has no federal law on election deepfakes — just a patchwork of state-level rules. The EU AI Act requires disclosure of manipulated content, but enforcement is slow. The core lesson from South Korea is that laws must be paired with platform accountability, detection infrastructure, and genuine media literacy among voters. Without all three, the gap between a deepfake going viral and anyone stopping it remains dangerously wide.
